The Method for Detecting Abnormal Actions in Industrial Robots Based on Robots Spatio-Temporal Graph Normalizing Flow

Rapid detection and response to abnormal movements of industrial robots can effectively reduce the risk of injury. Abnormal action detection requires modeling and distinguishing the complex and variable action patterns of industrial robots, which needs to take into account the spatio-temporal coupling and dynamic characteristics of industrial robots' actions, while efficient processing of spatio-temporal features and complex dynamics models requires a large amount of computational resources, which limits the accuracy and response speed of detection. To this end, we propose a robots spatio-temporal graph normalized flow (RSTG-NF) model for abnormal action detection of industrial robots. The model uses a Robots Spatio-Temporal Convolutional Network (RSTGC) to capture and differentiate the complex action relationships and motion patterns of industrial robots actions in both temporal and spatial dimensions. The frame-level action sequences are placed in a bi-directional mapping between the high-dimensional data space and the low-dimensional potential space through normalized flow, and the anomaly scores are evaluated based on the mapped data distributions for localization and detection of anomalies. Extensive experiments on a home-made six-axis industrial robots action dataset (IRADs) show an AUC score of 92.18% for frame-level detection and an inference speed of 79 FPS.